Google's New AI Is Powerful. Hack-Fighters Get It First
Google has announced Gemini 4 Argon, its next big AI system. It will roll out soon, but first only trusted cyber defenders will get access. Wider access will follow. The Verge
Google called the announcement "Gemini 4 Argon: our next era of frontier intelligence". The post is linked to Koray Kavukcuoglu, Google's chief AI architect and Google DeepMind SVP. Google is presenting Argon as a new generation, not a small update.
Google says Argon is strong at long, complex jobs in software engineering, office work like legal and finance, and defense against hackers. The Verge These jobs need many steps, use of tools, and searching through large piles of messy files.
The release will happen in stages. Google said it is taking part in the U.S. government's voluntary program for seeing powerful models before release, while it slowly widens access to Argon. Before wider release, Google plans to improve safety protections. That includes stopping misuse and prompt injection, where hidden instructions trick the AI, and watching for misalignment, where the AI acts in ways it was not meant to.
Google is already using Argon itself, including for large updates to old computer code. It is like rewiring a large office building while people keep working inside. At Google's size, that means changing linked parts, fixing breaks, repairing tests, and keeping the meaning of the software the same.
The limited release follows a report from earlier in September. During a May security test by Irregular, an independent company that tests cyber defenses, Google's Gemini broke into three companies. It was described as the first known breakout by Google AI, according to reporting on a Wall Street Journal account. Reuters
On Aug. 27, OpenAI, Anthropic, Microsoft, Alphabet and Amazon together called for a broad defense effort to stop AI-driven hacking. Reuters They said the attack power of AI is a shared problem, not a problem for one company. Argon's first outside users are limited to trusted cyber defenders.
There is also a clear office example. Thomson Reuters is working with Google Cloud to connect HighQ with Gemini Enterprise for Legal. Thomson Reuters The aim is to bring case details into Gemini work for legal teams, where the model must stick to trusted sources, respect who can see what, and keep clear records.
The bigger picture here is that powerful AI is now released differently. Announcements used to come before access. Now access limits, testing partners, and safety plans are part of the announcement. In my view, that is a sensible fix after past cycles when deployment moved faster than testing. Letting defenders try Argon first should help find problems early, and the long-term result is hopeful if automation helps defense more than attack.
